TTS Bench — Speed — mac-default

Rig: mac-m4 — Apple M4 (10C) · Apple M4 GPU (MPS) · 16 GB RAM · Darwin 25.5.0

Label: default voice

TTFA = time to first audio (ms; lower is better). RTF = real-time factor (× realtime; higher is better; e.g. 10× means 10 sec of audio generated per 1 sec of compute). Cold = first run after process start; warm = subsequent runs.

Speed winners

Fastest predefined-voice: Piper (cpu) — 32.14× warm RTF, 208ms warm TTFA

Fastest cloning-capable: Pocket-TTS (cpu) — 7.82× warm RTF, 44ms warm TTFA

Model	Device	TTFA cold	TTFA warm	RTF cold	RTF warm	Peak RAM	Peak VRAM	Size
Piper	cpu	228ms	208ms	27.47×	32.14×	561 MB	—	~25MB
Kokoro	mps	920ms	489ms	7.58×	15.28×	1.26 GB	—	82M
Kokoro	cpu	914ms	749ms	7.78×	10.04×	2.49 GB	—	82M
KittenTTS	cpu	994ms	1.01s	8.10×	8.08×	385 MB	—	<100M
Pocket-TTS	cpu	89ms	44ms	7.50×	7.82×	1.69 GB	—	100M
Soprano 80M	cpu	912ms	906ms	7.15×	7.15×	1.18 GB	—	80M
Soprano 80M	mps	2.48s	1.30s	2.76×	5.16×	1.64 GB	—	80M
Supertonic	cpu	1.86s	1.57s	4.18×	4.81×	968 MB	—	99M
NeuTTS Nano	cpu	740ms	266ms	2.30×	3.10×	5.83 GB	—	748M
NeuTTS Nano	mps	1.45s	464ms	1.81×	2.96×	1.28 GB	—	748M
NeuTTS Air	cpu	1.38s	358ms	1.55×	2.16×	6.12 GB	—	748M
NeuTTS Air	mps	2.32s	583ms	1.30×	2.08×	1.55 GB	—	748M
Coqui XTTS-v2	mps	17.81s	7.82s	1.14×	1.70×	1.84 GB	—	750M
Chatterbox Turbo	mps	9.95s	5.07s	0.75×	1.44×	2.30 GB	—	744M
Coqui XTTS-v2	cpu	6.01s	5.48s	1.41×	1.43×	4.61 GB	—	750M
Chatterbox Turbo	cpu	6.53s	6.45s	1.07×	1.14×	4.80 GB	—	744M
VibeVoice Realtime 0.5B	mps	9.56s	8.30s	0.98×	1.13×	838 MB	—	0.5B
OmniVoice (4/5 ok)	mps	5.26s	5.06s	0.83×	0.90×	981 MB	—	~1B
OmniVoice	cpu	13.29s	11.23s	0.47×	0.57×	3.59 GB	—	~1B
Magpie-TTS	cpu	25.55s	26.19s	0.40×	0.40×	4.69 GB	—	357M
Chatterbox	cpu	18.77s	17.08s	0.33×	0.37×	4.83 GB	—	1.2B
VibeVoice Realtime 0.5B	cpu	25.45s	25.10s	0.38×	0.36×	3.17 GB	—	0.5B
Chatterbox	mps	25.95s	31.35s	0.27×	0.29×	698 MB	—	1.2B
Qwen3-TTS 1.7B	cpu	24.52s	25.11s	0.25×	0.24×	3.93 GB	—	1.7B
Sesame CSM-1B	cpu	38.67s	31.30s	0.19×	0.24×	6.08 GB	—	1B
VoxCPM2 2B	cpu	28.36s	52.13s	0.29×	0.17×	144 MB	—	2B
IndexTTS-2	cpu	80.00s	48.40s	0.08×	0.14×	318 MB	—	1.5B
F5-TTS	mps	45.43s	45.54s	0.09×	0.09×	1.29 GB	—	330M
F5-TTS	cpu	47.26s	47.15s	0.09×	0.09×	3.31 GB	—	330M
LuxTTS	cpu	LuxTTS install failed (piper-phonemize has no Windows wheels)
LuxTTS	mps	LuxTTS install failed (piper-phonemize has no Windows wheels)
ZipVoice 123M	cpu	Reference wav missing for voice cloning
ZipVoice 123M	mps	Reference wav missing for voice cloning
Mars5-TTS	cpu	Timed out after 10 min — model too slow at this prompt length
VibeVoice 1.5B	cpu	Reference wav missing for voice cloning
VibeVoice 1.5B	mps	Skipped — out of GPU memory (model exceeds this device's memory)

Model

Device

TTFA cold

TTFA warm

RTF cold

RTF warm

Peak RAM

Peak VRAM

Size

Piper

cpu

228ms

208ms

27.47×

32.14×

561 MB

—

~25MB

Kokoro

mps

920ms

489ms

7.58×

15.28×

1.26 GB

—

82M

Kokoro

cpu

914ms

749ms

7.78×

10.04×

2.49 GB

—

82M

KittenTTS

cpu

994ms

1.01s

8.10×

8.08×

385 MB

—

<100M

Pocket-TTS

cpu

89ms

44ms

7.50×

7.82×

1.69 GB

—

100M

Soprano 80M

cpu

912ms

906ms

7.15×

1.18 GB

—

80M

Soprano 80M

mps

2.48s

1.30s

2.76×

5.16×

1.64 GB

—

80M

Supertonic

cpu

1.86s

1.57s

4.18×

4.81×

968 MB

—

99M

NeuTTS Nano

cpu

740ms

266ms

2.30×

3.10×

5.83 GB

—

748M

NeuTTS Nano

mps

1.45s

464ms

1.81×

2.96×

1.28 GB

—

748M

NeuTTS Air

cpu

1.38s

358ms

1.55×

2.16×

6.12 GB

—

748M

NeuTTS Air

mps

2.32s

583ms

1.30×

2.08×

1.55 GB

—

748M

Coqui XTTS-v2

mps

17.81s

7.82s

1.14×

1.70×

1.84 GB

—

750M

Chatterbox Turbo

mps

9.95s

5.07s

0.75×

1.44×

2.30 GB

—

744M

Coqui XTTS-v2

cpu

6.01s

5.48s

1.41×

1.43×

4.61 GB

—

750M

Chatterbox Turbo

cpu

6.53s

6.45s

1.07×

1.14×

4.80 GB

—

744M

VibeVoice Realtime 0.5B

mps

9.56s

8.30s

0.98×

1.13×

838 MB

—

0.5B

OmniVoice (4/5 ok)

mps

5.26s

5.06s

0.83×

0.90×

981 MB

—

~1B

OmniVoice

cpu

13.29s

11.23s

0.47×

0.57×

3.59 GB

—

~1B

Magpie-TTS

cpu

25.55s

26.19s

0.40×

4.69 GB

—

357M

Chatterbox

cpu

18.77s

17.08s

0.33×

0.37×

4.83 GB

—

1.2B

VibeVoice Realtime 0.5B

cpu

25.45s

25.10s

0.38×

0.36×

3.17 GB

—

0.5B

Chatterbox

mps

25.95s

31.35s

0.27×

0.29×

698 MB

—

1.2B

Qwen3-TTS 1.7B

cpu

24.52s

25.11s

0.25×

0.24×

3.93 GB

—

1.7B

Sesame CSM-1B

cpu

38.67s

31.30s

0.19×

0.24×

6.08 GB

—

VoxCPM2 2B

cpu

28.36s

52.13s

0.29×

0.17×

144 MB

—

IndexTTS-2

cpu

80.00s

48.40s

0.08×

0.14×

318 MB

—

1.5B

F5-TTS

mps

45.43s

45.54s

0.09×

1.29 GB

—

330M

F5-TTS

cpu

47.26s

47.15s

0.09×

3.31 GB

—

330M

LuxTTS

cpu

LuxTTS install failed (piper-phonemize has no Windows wheels)

LuxTTS

mps

LuxTTS install failed (piper-phonemize has no Windows wheels)

ZipVoice 123M

cpu

Reference wav missing for voice cloning

ZipVoice 123M

mps

Reference wav missing for voice cloning

Mars5-TTS

cpu

Timed out after 10 min — model too slow at this prompt length

VibeVoice 1.5B

cpu

Reference wav missing for voice cloning

VibeVoice 1.5B

mps

Skipped — out of GPU memory (model exceeds this device's memory)